We build continental-scale models to predict biomass using millions of pre-filtered, spaceborne and airborne LiDAR estimates. Our models are validated with extensive airborne LiDAR and field data. This allows to capture the full range of variation in carbon density over the region of interest.

We derive features from state-of-the-art fusion of active and passive Earth observation and ancillary data. We leverage the unique capability of each sensor to provide measurements of the state of natural ecosystems. This technique allows us to overcome limitations where data gaps exist in time and space, and produces reliable inputs to our machine learning model for wall-to-wall biomass estimates.

We build machine-learning models anchored in state-of-the-art data science. Our models are based on highly curated features that, when combined with LiDAR measurements, generate high spatial resolution time series of carbon stock and change over long periods. Our continental scale models are capable of producing on-demand, wall-to-wall maps of above-ground carbon stocks and change, at annual time steps with a spatial resolution of 10 or 30 meters.

Our cloud-based pipeline processes annual carbon stock estimates since the year 2000 at each grid cell using Bayesian time series methods to remove noise in the time series and confidently identify statistically significant changes and trends in carbon stocks across time, with quantified uncertainty at the pixel level.